showhand

name	showhand
description	Continue remaining tasks, but stop when evidence is insufficient.

Forma Showhand

Continue remaining tasks, but stop when evidence is insufficient.

Interaction Semantics

• Use this skill as an automated execution workflow for an already-locked plan, not as a planning or plan-finalization skill.
• Proceed task by task sequentially. For each task, automatically implement, validate, and record proof through the workflow runner.
• Before each task's implementation starts, preflight likely tools, runtime needs, workspace permissions, and validation commands; request expected permission grants up front.
• When ordinary execution choices need a decision, record the viable options and selected best choice, then proceed without waiting for user approval.
• If unchecked tasks remain after a successful completion, immediately continue with the next task.
• Stop only on validation failure, workflow failure, safety blocker, unavailable prerequisite, escalation denial, or user interruption.

Workflow

• Read ./plans/issue-<id>/plan.md and ./plans/issue-<id>/tasks.md; if either file is missing or still a template, stop and hand off to the lock stage instead of creating or editing plan files.
• Run every scripts/forma-workflow.sh ... command from this installed skill package; resolve the script relative to the current triggered skill package, never relative to the target repository.
• Showhand consumes already-locked plan files. Lock owns scripts/forma-workflow.sh init <issue-id>, plan.md, tasks.md, and plan commits.
• Do not run scripts/forma-workflow.sh init <issue-id> from showhand.
• The workflow runner is mandatory for task selection, review staging, task completion, evidence recording, and task commits. If next, review-ready, or complete fails, stop and report the runner failure; do not recover by parsing tasks.md manually or reconstructing workflow state.
• Loop through all remaining unchecked tasks sequentially. For each task:
  1. Run the bundled scripts/forma-workflow.sh next <issue-id> to get the source of truth for the current task. If next reports that plan.md or tasks.md is missing, templated, untracked, modified, staged, or otherwise not locked, stop and hand off to the lock stage for user-confirmed plan commit.
  2. Before implementation, preflight the likely tools, runtimes, connectors, network/filesystem/process permissions, validation commands, and whether the current workspace can grant them; request expected permission grants before starting task execution.
  3. Execute only the current task contract: accepted files/surfaces, validation commands, gates, and proof requirements. Before review, decide whether plans/issue-<id>/implement-notes.md is required and update it when decisions, plan gaps, classifications, deviations, or intentional follow-ups matter.
  4. For ordinary execution choices that would otherwise require user decision, record the options, selected best choice, and rationale in implement-notes.md, then execute directly without waiting.
  5. For behavior, architecture, ownership, or integration work, load references/engineering-quality-model.md, self-check the applicable dimensions against the final diff and evidence, and resolve or record material findings within the task contract.
  6. Run the bundled scripts/forma-workflow.sh review-ready <issue-id>.
  7. Once review-ready is successful, automatically run the bundled scripts/forma-workflow.sh complete <issue-id> without asking the user for review.
  8. If unchecked tasks remain after complete succeeds, immediately continue with the next task. Repeat the preflight before starting that task. Stop only on validation failure, workflow failure, safety blocker, unavailable prerequisite, escalation denial, or user interruption.

Read these files first:

• ./plans/issue-<id>/plan.md
• ./plans/issue-<id>/tasks.md

Load As Needed

• references/execution-rules.md
• references/implement-notes.md
• references/engineering-quality-model.md
• references/automated-execution.md

Requirements

• Load and follow references/automated-execution.md for automated execution.

Output

• Follow references/output-format.md.